Robust modeling for multivariate calibration transfer by the successive projections algorithm

The use of the Successive Projections Algorithm (VSPA) to select variables for building robust transferable Multiple Linear Regression (MLR) models is investigated. The robustness requirement is explicitly taken into account by introducing some spectra acquired by the slave instrument in the validation set that guides VSPA selection. The transfer samples are selected by the classic Kennard–Stone (KS) algorithm or a modified version of VSPA (SSPA) that operates across the rows of the instrumental response matrix, instead of the columns. The proposed approach was tested in two data sets, each set consisting of spectra obtained by infrared spectrometry in two different instruments. The first data set consists of gasoline spectra, which are employed to predict the distillation temperature at which 90% of the sample has evaporated (T 90%). The second data set consists of corn spectra, which are employed for moisture determination. The robust MLR models were compared to a PLS model employing Piecewise Direct Standardization (PDS) to correct the slave spectra. In both data sets, the mean prediction errors at the slave instrument for the robust VSPA-MLR and the PLS-PDS models were comparable, and slightly better for VSPA-MLR. The proposed approach appears to be a valid alternative to the commonly used PLS-PDS technique.